Subprocessors
Last updated: 2026-06-18
ilion uses a small set of third-party processors to deliver the service. This page lists every one of them, what it does, and — critically — that most only ever receive data when your organization explicitly connects the corresponding integration.
Core service — always involved
| Subprocessor | Purpose | Processing location |
|---|---|---|
| Supabase | Database, authentication, file storage | EU (AWS eu-west-1, Ireland) |
| Vercel | Application hosting and delivery | EU (functions pinned to Dublin) / global CDN |
| Anthropic | AI model inference for agents and chat | US/EU (API) |
| Stripe | Payment processing and billing | US/EU |
| Resend | Transactional and campaign email delivery | US/EU |
| Trigger.dev | Background job execution for agent runs | US/EU |
Optional integrations — only when you connect them
These processors receive no customer data unless an administrator or user of your organization explicitly enables the corresponding feature. Disconnecting the integration stops the data flow.
| Subprocessor | Purpose | Processing location | When data flows |
|---|---|---|---|
| Gmail and Google Calendar sync | Global | Only when a user connects their Google account | |
| Microsoft | Outlook email and calendar sync | Global | Only when a user connects their Microsoft account |
| Twilio | Telephony — calls and SMS | US/EU | Only when the organization connects telephony |
| Deepgram | Call transcription (speech-to-text) | EU endpoint | Only when call recording/transcription is enabled |
| OpenAI | Fallback call transcription (speech-to-text) | US | Only when call transcription is enabled and the primary provider is unavailable |
| Recall.ai | Managed meeting notetaker (bot join + recording) | US/EU | Only when the organization enables the meeting notetaker |
| Tavily | Web search for lead/account enrichment | US | Only when enrichment agents run web research |
| Hunter | Business contact discovery for enrichment | EU | Only when enrichment/prospecting is used |
AI processing, stated plainly
Agent and chat features send relevant record context to Anthropic's API to generate output. Anthropic does not train models on this API data. AI-generated actions that change your data go through an approval queue — see the security page for the full model.
Changes to this list
We will update this page before adding or replacing a subprocessor that processes customer personal data. Customers with a Data Processing Agreement in place will be notified per its terms and may object as described there — see the DPA.